A dynamic, contamination‑free benchmark for evaluating LLMs on complex, real‑world text‑to‑SQL tasks.
🌐 Website/Leaderboard • 📄 Paper (coming soon) • 💻 GitHub • 🗄️ LiveSQLBench-Base-Lite • 🗄️ LiveSQLBench-Large-v1 • 🗄️ Bird-Interact (ICLR 2026 Oral)
Maintained by the 🦜 BIRD Team @ HKU & ☁️ Google Cloud
LiveSQLBench (BIRD-SQL Pro v0.5) is a contamination-free, continuously evolving benchmark designed to… See the full description on the dataset page:
https://huggingface.co/datasets/birdsql/livesqlbench-base-full-v1.
